Thomas Dohmke is leaving GitHub to become a founder again, leading him to start a new company called Entire. He will stay at GitHub through the end of 2025 to help guide the company transition. His departure removes the single leader role from the platform, as GitHub and its leadership team now work as part of Microsoft’s CoreAI organization. This division, which Microsoft established in January 2025, is led by Jay Parikh. A memo from Parikh indicates that GitHub executives report to Julia Liuson, a former head of Microsoft’s developer division. This reorganization follows the 2018 acquisition of GitHub by Microsoft for $7.5 billion.
Organizational changes at Microsoft
The decision to fold GitHub into CoreAI caused internal disaffection and a departure of senior leadership. At least 11 of the current 30 employees are former GitHub veterans who left after the reorganization. Julia Liuson announced her resignation, and former Chief Revenue Officer Elizabeth Pemmerl also left. Jared Palmer, who joined as Senior Vice President in October 2025, moved to Xbox. Many employees express that the company lost its independence and direction. The company no longer has a single leader or CEO, as responsibility for GitHub aligns with the CoreAI leadership team.
The workflow of Copilot Workspace
Copilot Workspace uses GPT-4 Turbo to implement code from natural language instructions. You know the basics of how an autocomplete tool works, so I will focus on the agentic workflow. The process moves through specific stages to turn a concept into a reality. A developer provides a task, which is a natural language description of the desired change. The system then generates a specification, which maps the current state of the codebase and the desired state. The user can edit these lists to correct the system understanding or refine the requirements.
Once the user accepts the specification, the tool generates a plan. This plan lists every file the agent intends to create, modify, or delete, and provides a list of actions for each file. The user can edit the plan to change specific instructions or the list of files affected. After the user approves the plan, the implementation stage begins. The tool writes the corresponding code, and the developer can then see a diff of each file.
| Feature | Detail |
|---|---|
| Underlying Model | GPT-4 Turbo |
| Primary Interface | Natural language specification and plan editing |
| Integration | Codespaces, terminal, and browser |
| Capabilities | Task planning, code implementation, and diff editing |
| Execution | Live sync with Codespaces for terminal and dependency management |
A developer once used the tool to replace a tribal tattoo logo with a new logo from a designer named Keshav. The task included the specific file to modify, where the logo appears, and the file name. The system read the codebase to create a specification of the current and desired states. The developer then reviewed the plan, which listed the changes for each file. The tool wrote the code, and the developer checked the diff before creating a pull request.
The desktop application as a control plane
The Copilot desktop application functions as a control plane for autonomous coding efforts. It provides a dedicated space for sessions, branches, worktrees, terminals, browsers, and diffs to manage the messy business of automated code changes within familiar software delivery controls. The application is available on Windows, macOS, and Linux. A developer can start a session from an issue, a pull request, or a custom prompt. Each session runs on its own branch and worktree, so multiple streams of AI-generated changes proceed without interfering with the current working directory.
The tool allows for parallel sessions where one agent handles a test failure, another updates documentation, and a third explores a refactor. This design targets developers who need to supervise multiple streams of work. The application also includes cloud automations. These allow users to schedule recurring agent work in the cloud. This means Copilot sessions do not require a developer to be awake or connected to the internet. Scheduled agents can refresh dependencies, update generated files, or triage low-priority issues.
Model brokerage and the MCP registry
GitHub serves as a broker for multiple models, including those from Anthropic, Google, and OpenAI. The platform allows developers to choose the specific model behind each session based on the task. A small, fast model might handle boilerplate, while a stronger reasoning model handles cross-repository refactors. The new MCP Registry provides a centralized directory for Model Context Protocol servers. This registry organizes servers by GitHub stars and community activity.
The registry allows for one-click installation of servers in VS Code. GitHub is working with Anthropic and the MCP Steering Committee to build an open-source registry. Developers can self-publish to these registries to ensure servers appear in both locations. GitHub also partners with Figma, Postman, HashiCorp, and Dynatrace to set quality standards for these integrations. This setup connects Copilot sessions to internal documentation, build systems, issue trackers, and cloud environments.
Infrastructure stability and security vulnerabilities
GitHub infrastructure experiences frequent outages due to high volumes of pull requests and commits. CTO Vladimir Fedorov stated that the team prioritizes availability and capacity over new features. He is working to reduce non-essential work and improve caching. The team is also working to isolate critical services to eliminate single points of failure. Fedorov began a project to migrate GitHub onto Azure servers to address data center capacity bottlenecks. The complexity of MySQL clusters makes outage risks during this migration difficult to eliminate.
Security vulnerabilities also present a challenge for the platform. In March, security firm Wiz Research found a critical vulnerability in the internal git infrastructure using AI models. A more severe breach occurred when 3,800 internal GitHub repositories were compromised. This breach originated from an employee installing a VS Code extension that contained malicious code. Microsoft insiders said that VS Code frequently pushes extension installation prompts onto users.
Competition and the billing model
The market for AI coding assistants includes competitors like Cursor, Claude Code, and Dohmke’s company, Entire. These competitors view the current market as an opportunity to reshape the landscape. GitHub is implementing a usage-based billing model for Copilot to manage these pressures. This model restricts each subscription tier to a monthly pool of AI credits, and if a user exceeds this limit, the service cuts off instead of downgrading to a lower-capability model. This change causes friction among developers who prefer unlimited testing.
The industry is also seeing a shift in how developers view AI agents. Some users find that Copilot Workspace is better at capturing a thought process than a simple chat interface. Others find that the task of writing requirements is still difficult because the user may not have a perfect idea of the feature. This difficulty makes it hard to discern if the AI is making the correct design decisions. The new billing thresholds and the rise of AI-native tools like Cursor create pressure on the GitHub ecosystem.
The scale of the developer ecosystem
GitHub has grown to include over 150 million registered developers. This is a significant increase from the 73 million developers recorded in 2021. The platform hosts more than 1 billion repositories and forks. AI projects on the platform have doubled in the last year alone. Copilot has over 20 million users.
The platform maintains its presence in companies of all sizes. GitHub Copilot began as a tool for autocompletion but has expanded to include conversational coding and full-stack app creation with GitHub Spark. The company continues to expand its reach in the EU and Australia. Will the integration into CoreAI eventually change the way the community interacts with open source repositories? GitHub remains a major platform for the global software industry.
